Point-Voxel CNN for Efficient 3D Deep Learning

   Zhijian Liu         Haotian Tang             Yujun Lin      Song Han
     MIT         Shanghai Jiao Tong University         MIT         MIT

                         Abstract
     We present Point-Voxel CNN (PVCNN) for efficient, fast 3D deep learning. Previ-
     ous work processes 3D data using either voxel-based or point-based NN models.
     However, both approaches are computationally inefficient. The computation cost
     and memory foot ...